CI test-phi-3-mini-runner-linux failing as a result of transformers v4.54.0

[Conarnar]

CI test-phi-3-mini-runner-linux fails with the following error:

Traceback (most recent call last):
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/envs/py_3.10/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/executorch/examples/models/phi-3-mini/export_phi-3-mini.py", line 168, in <module>
    main()
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/executorch/examples/models/phi-3-mini/export_phi-3-mini.py", line 164, in main
    export(parser.parse_args())
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/executorch/examples/models/phi-3-mini/export_phi-3-mini.py", line 79, in export
    exportable_module = TorchExportableModuleForDecoderOnlyLM(
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/transformers/integrations/executorch.py", line 67, in __init__
    self.model = TorchExportableModuleWithStaticCache(model)
  File "/opt/conda/envs/py_3.10/lib/python3.10/site-packages/transformers/integrations/executorch.py", line 293, in __init__
    max_batch_size=self.model.generation_config.cache_config.get("batch_size"),
AttributeError: 'StaticCacheConfig' object has no attribute 'get'

This appears to be caused by the release of transformers v4.54.0, specifically as a result of the lines changed in src/transformers/integrations/executorch.py.

[guangy10]

Seems like due some breaking changes in the upstream huggingface/transformers#39106. Currently in optimum-executorch the pinned version is 4.53.4. That version should be good

[guangy10]

I see the pinned version is downgraded to 4.53.1 to unbreak the CI already. #12900

[manueldeprada]

We have refactored Caches in transformers so that they are defined per-layer and also simplified the configs.

The config objects like StaticCacheConfig are deprecated and a dict is encouraged instead. If you change that, it should work like before :)